Upgrade to LLVM.jl 10

[maleadt]

Ports Metal.jl to LLVM.jl 10.

Main changes:

• Mechanical port: compat, vocabulary imports, properties instead of accessors, and explicit insertion points. The threadgroup memory, RNG state and @mtlprintf generators use @llvmgenerated.
• No more raw C API: scoped enums, get/get! on the module views, llvm.memcpy recognized as an intrinsic rather than by its name prefix, and strip_pointer_casts instead of hand-written pointer walks. @mtlprintf declares the varargs and lifetime intrinsics through Intrinsic, which handles their changes in LLVM 19 and 22. The volumerhs benchmark uses @llvmgenerated.
• IR ownership: the IR that GPUCompiler returns is disposed of inside the JuliaContext block, also when lowering to AIR fails.

Fixes:

• finish_module! overwrote the debug location of the kernel's first instruction with a line-0 location, and the AIR downgrades copied a call's location onto itself.
• The simdgroup-matrix downgrade passed non-constant vectors to LLVMGetAggregateElement, which asserts on LLVM builds with assertions.

Tested on an M1 on Julia 1.10, 1.12 and 1.13, including GPUArrays' tests. Results match main: on 1.10, the two fence checks in synchronization.jl fail as they do on main, and the gtk example needs a window server.

Requires GPUCompiler 2.11 (JuliaGPU/GPUCompiler.jl#974), GPUToolbox 3.3.3 (JuliaGPU/GPUToolbox.jl#28), UnsafeAtomics 0.3.3 (JuliaConcurrent/UnsafeAtomics.jl#31) and GPUArrays 11.5.16 (JuliaGPU/GPUArrays.jl#801).

---

Disclaimer: this PR is AI-assisted and has not been reviewed in detail. Tests pass locally, so it should be a good starting point for a maintainer to complete the upgrade.

[github-actions]

Metal Benchmarks

Details

Benchmark suite	Current: 440e539	Previous: 6eb4996	Ratio
array/accumulate/Float32/1d	388542 ns	393500 ns	0.99
array/accumulate/Float32/dims=1	366875 ns	368750 ns	0.99
array/accumulate/Float32/dims=1L	8847042 ns	8879000 ns	1.00
array/accumulate/Float32/dims=2	431500 ns	436667 ns	0.99
array/accumulate/Float32/dims=2L	2640958 ns	2760209 ns	0.96
array/accumulate/Int64/1d	842625 ns	849667 ns	0.99
array/accumulate/Int64/dims=1	908917 ns	900750 ns	1.01
array/accumulate/Int64/dims=1L	9553709 ns	9583625 ns	1.00
array/accumulate/Int64/dims=2	1210084 ns	1215250 ns	1.00
array/accumulate/Int64/dims=2L	6557542 ns	6538208 ns	1.00
array/broadcast	227958 ns	237708 ns	0.96
array/construct	2292 ns	2292 ns	1
array/permutedims/2d	458666 ns	444000 ns	1.03
array/permutedims/3d	1030542 ns	1029208 ns	1.00
array/permutedims/4d	1115167 ns	1119959 ns	1.00
array/private/copy	229459 ns	228417 ns	1.00
array/private/copyto!/cpu_to_gpu	213709 ns	213417 ns	1.00
array/private/copyto!/gpu_to_cpu	216458 ns	210208 ns	1.03
array/private/copyto!/gpu_to_gpu	220542 ns	219250 ns	1.01
array/private/iteration/findall/bool	1061667 ns	1056458 ns	1.00
array/private/iteration/findall/int	1224875 ns	1217917 ns	1.01
array/private/iteration/findfirst/bool	1155667 ns	1154084 ns	1.00
array/private/iteration/findfirst/int	1161042 ns	1158833 ns	1.00
array/private/iteration/findmin/1d	1193458 ns	1191875 ns	1.00
array/private/iteration/findmin/2d	1028542 ns	1022417 ns	1.01
array/private/iteration/logical	1667166 ns	1663292 ns	1.00
array/private/iteration/scalar	1406500 ns	1380333 ns	1.02
array/random/rand/Float32	419625 ns	418875 ns	1.00
array/random/rand/Int64	493167 ns	494333 ns	1.00
array/random/rand!/Float32	379292 ns	403916 ns	0.94
array/random/rand!/Int64	427375 ns	391417 ns	1.09
array/random/randn/Float32	393833 ns	384167 ns	1.03
array/random/randn!/Float32	377375 ns	368417 ns	1.02
array/reductions/mapreduce/Float32/1d	450709 ns	438709 ns	1.03
array/reductions/mapreduce/Float32/dims=1	356417 ns	290625 ns	1.23
array/reductions/mapreduce/Float32/dims=1L	617459 ns	618333 ns	1.00
array/reductions/mapreduce/Float32/dims=2	358583 ns	349250 ns	1.03
array/reductions/mapreduce/Float32/dims=2L	899166 ns	1094500 ns	0.82
array/reductions/mapreduce/Int64/1d	621042 ns	636292 ns	0.98
array/reductions/mapreduce/Int64/dims=1	637125 ns	631500 ns	1.01
array/reductions/mapreduce/Int64/dims=1L	1007583 ns	1010458 ns	1.00
array/reductions/mapreduce/Int64/dims=2	791875 ns	785625 ns	1.01
array/reductions/mapreduce/Int64/dims=2L	2191167 ns	2189584 ns	1.00
array/reductions/reduce/Float32/1d	449625 ns	449375 ns	1.00
array/reductions/reduce/Float32/dims=1	360708 ns	355666 ns	1.01
array/reductions/reduce/Float32/dims=1L	616667 ns	607917 ns	1.01
array/reductions/reduce/Float32/dims=2	246291 ns	249042 ns	0.99
array/reductions/reduce/Float32/dims=2L	463250 ns	473250 ns	0.98
array/reductions/reduce/Int64/1d	638792 ns	629875 ns	1.01
array/reductions/reduce/Int64/dims=1	637084 ns	637833 ns	1.00
array/reductions/reduce/Int64/dims=1L	1018959 ns	1005875 ns	1.01
array/reductions/reduce/Int64/dims=2	256791 ns	268709 ns	0.96
array/reductions/reduce/Int64/dims=2L	672542 ns	666541 ns	1.01
array/shared/copy	129917 ns	131292 ns	0.99
array/shared/copyto!/cpu_to_gpu	38334 ns	38042 ns	1.01
array/shared/copyto!/gpu_to_cpu	37959 ns	37292 ns	1.02
array/shared/copyto!/gpu_to_gpu	38250 ns	37625 ns	1.02
array/shared/iteration/findall/bool	1062750 ns	1045500 ns	1.02
array/shared/iteration/findall/int	1220291 ns	1221500 ns	1.00
array/shared/iteration/findfirst/bool	981459 ns	977500 ns	1.00
array/shared/iteration/findfirst/int	991709 ns	986500 ns	1.01
array/shared/iteration/findmin/1d	1054333 ns	1066708 ns	0.99
array/shared/iteration/findmin/2d	1033375 ns	1030959 ns	1.00
array/shared/iteration/logical	1525292 ns	1530416 ns	1.00
array/shared/iteration/scalar	4666.714285714285 ns	4875 ns	0.96
array/sorting/1d	2108250 ns	2093791 ns	1.01
array/sorting/2d	8431166 ns	8467208 ns	1.00
integration/byval/reference	1105250 ns	1123250 ns	0.98
integration/byval/slices=1	1127125 ns	1116541 ns	1.01
integration/byval/slices=2	2028750 ns	2019584 ns	1.00
integration/byval/slices=3	6846292 ns	6809084 ns	1.01
integration/metaldevrt	400042 ns	392417 ns	1.02
kernel/indexing	216708 ns	208833 ns	1.04
kernel/indexing_checked	394500 ns	393834 ns	1.00
kernel/launch	2074 ns	2111.1111111111113 ns	0.98
kernel/rand	402084 ns	403041 ns	1.00
latency/import	1860023458 ns	1821483458 ns	1.02
latency/precompile	33629446292 ns	32771377958 ns	1.03
latency/ttfp	2287592542 ns	2276943625 ns	1.00
metal/synchronization/context	528.7315789473685 ns	518.7643979057592 ns	1.02
metal/synchronization/stream	439.8131313131313 ns	459.39086294416245 ns	0.96

This comment was automatically generated by workflow using github-action-benchmark.

[codecov]

Codecov Report

❌ Patch coverage is 98.52941% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 87.23%. Comparing base (6eb4996) to head (440e539).

Files with missing lines	Patch %	Lines
src/compiler/compilation.jl	97.22%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #997      +/-   ##
==========================================
+ Coverage   87.09%   87.23%   +0.14%     
==========================================
  Files          92       92              
  Lines        6747     6666      -81     
==========================================
- Hits         5876     5815      -61     
+ Misses        871      851      -20     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.